Search

The interaction of land with the atmosphere is sensitive to soil moisture (W). Evapotranspiration (ET) reacts to soil moisture in a nonlinear way, f(W), as soils dry from saturation to wilt point. This nonlinear behavior and the fact that soil moisture varies on scales as small as 1–10 m in nature, while numerical general circulation models (GCMs) have grid cell sizes on the order of 1 to 100s of kilometers, makes the calculation of grid cell-average ET problematic. It is impractical to simulate the land in GCMs on the small scales seen in nature, so techniques have been developed to represent subgrid scale heterogeneity, including: (1) statistical-dynamical representations of grid subelements of varying wetness, (2) relaxation of f(W), (3) moderating f(W) with approximations of catchment hydrology, (4) “tiling” the landscape into vegetation types, and (5) hyperresolution. Here we present an alternative method for representing subgrid variability in W, one proven in a conceptual framework where landscape-scale W is represented as a series of “Bins” of increasing wetness from dry to saturated. The grid cell-level f(W) is defined by the integral of the fractional area of the wetness bins and the value of f(W) associated with each. This approach accounts for the spatiotemporal dynamics of W. We implemented this approach in the SiB3 land surface parameterization and then evaluated its performance against a control, which assumes a horizontally uniform field of W. We demonstrate that the Bins method, with a physical basis, attenuates unrealistic jumps in model state and ET seen in the control runs.

Cloud fraction (CF) significantly affects the short‐ and long‐wave radiation. Its realistic representation in general circulation models (GCMs) still poses great challenges in modeling the atmosphere. Here, we present a neural network‐based (NN‐based) diagnostic scheme that uses the grid‐mean temperature, pressure, liquid and ice water mixing ratios, and relative humidity to simulate the sub‐grid CF. The scheme, trained using CloudSat data with explicit consideration of grid sizes, realistically simulates the observed CF with a correlation coefficient >0.9 for liquid‐, mixed‐, and ice‐phase clouds. The scheme also captures the observed non‐monotonic relationship between CF and relative humidity and is computationally efficient, and robust for GCMs with a variety of horizontal and vertical resolutions. For illustrative purposes, we conducted comparative analyses of the 2006–2019 climatological‐mean cloud fractions among CloudSat, and simulations from the NN‐based scheme and the Xu‐Randall scheme (optimized the same way as the NN‐based scheme). The NN‐based scheme improves not only the spatial distribution of the total CF but also the cloud vertical structure. For example, the biases of too‐many high‐level clouds over the tropics and too‐many low‐level clouds over regions around 60°S and 60°N in the Xu‐Randall scheme are significantly reduced. These improvements are also found to be insensitive to the spatio‐temporal variability of large‐scale meteorology conditions, implying that the scheme can be used in different climate regimes. Plain Language Summary: Cloud fraction (CF) is crucial to cloud climate effects of both reflecting shortwave radiation and absorbing/emitting longwave radiation. However, the simulation of CF in general circulation models (GCMs) has been difficult, because most clouds are smaller than the typical scales of GCM grids and cannot be resolved by the grid‐scale physics, while the physical understanding of sub‐grid processes is still inadequate. Thus, this study uses a data‐driven approach, that is, a neural network, to parameterize the sub‐grid CF in climate models. The database for training and evaluating this NN‐based scheme is obtained by upscaling the CloudSat (quasi‐) observational data and emulating the GCM "grid‐mean" properties required for cloud‐fraction parameterization, minimizing the data‐oriented biases. Moreover, the effects of the GCM horizontal and vertical grid sizes are both considered in the network, increasing the scheme adaptivity for use in GCMs with different resolutions. Results show that the NN‐based scheme correctly predicts observed features of cloud‐fraction variation with cloud condensate content and relative humidity for clouds of different phases and better predicts total cloud‐fraction spatial distribution and cloud vertical structure than the conventional Xu‐Randall scheme. This suggests that the NN‐based scheme has the potential to reduce the biases of cloud radiative effects existing in current GCMs. Key Points: A neural network‐based scheme for parameterizing sub‐grid cloud fraction in climate models is developed using the CloudSat dataThe scheme considers the effects of both horizontal and vertical grid sizes on cloud‐fraction parameterizationThe scheme better predicts total cloud‐fraction spatial distribution and cloud vertical structure than the Xu‐Randall scheme [ABSTRACT FROM AUTHOR]

Abstract The interaction of land with the atmosphere is sensitive to soil moisture (W). Evapotranspiration (ET) reacts to soil moisture in a nonlinear way, f(W), as soils dry from saturation to wilt point. This nonlinear behavior and the fact that soil moisture varies on scales as small as 1–10 m in nature, while numerical general circulation models (GCMs) have grid cell sizes on the order of 1 to 100s of kilometers, makes the calculation of grid cell‐average ET problematic. It is impractical to simulate the land in GCMs on the small scales seen in nature, so techniques have been developed to represent subgrid scale heterogeneity, including: (1) statistical‐dynamical representations of grid subelements of varying wetness, (2) relaxation of f(W), (3) moderating f(W) with approximations of catchment hydrology, (4) “tiling” the landscape into vegetation types, and (5) hyperresolution. Here we present an alternative method for representing subgrid variability in W, one proven in a conceptual framework where landscape‐scale W is represented as a series of “Bins” of increasing wetness from dry to saturated. The grid cell‐level f(W) is defined by the integral of the fractional area of the wetness bins and the value of f(W) associated with each. This approach accounts for the spatiotemporal dynamics of W. We implemented this approach in the SiB3 land surface parameterization and then evaluated its performance against a control, which assumes a horizontally uniform field of W. We demonstrate that the Bins method, with a physical basis, attenuates unrealistic jumps in model state and ET seen in the control runs.

In this study, it is shown that CMIP5 global climate models (GCMs) that convert supercooled water to ice at relatively warm temperatures tend to have a greater mean-state cloud fraction and more negative cloud feedback in the middle and high latitude Southern Hemisphere. We investigate possible reasons for these relationships by analyzing the mixed-phase parameterizations in 26 GCMs. The atmospheric temperature where ice and liquid are equally prevalent (T5050) is used to characterize the mixed-phase parameterization in each GCM. Liquid clouds have a higher albedo than ice clouds, so, all else being equal, models with more supercooled liquid water would also have a higher planetary albedo. The lower cloud fraction in these models compensates the higher cloud reflectivity and results in clouds that reflect shortwave radiation (SW) in reasonable agreement with observations, but gives clouds that are too bright and too few. The temperature at which supercooled liquid can remain unfrozen is strongly anti-correlated with cloud fraction in the climate mean state across the model ensemble, but we know of no robust physical mechanism to explain this behavior, especially because this anti-correlation extends through the subtropics. A set of perturbed physics simulations with the Community Atmospheric Model Version 4 (CAM4) shows that, if its temperature-dependent phase partitioning is varied and the critical relative humidity for cloud formation in each model run is also tuned to bring reflected SW into agreement with observations, then cloud fraction increases and liquid water path (LWP) decreases with T5050, as in the CMIP5 ensemble. [ABSTRACT FROM AUTHOR]

The responses of tropical anvil cloud and low‐level cloud to a warming climate are among the largest sources of uncertainty in our estimates of climate sensitivity. However, most research on cloud feedbacks relies on either global climate models with parameterized convection, which do not explicitly represent small‐scale convective processes, or small‐domain models, which cannot directly simulate large‐scale circulations. We investigate how self‐aggregation, the spontaneous clumping of convection in idealized numerical models, depends on cloud‐radiative interactions with different cloud types, sea surface temperatures (SSTs), and stages of aggregation in simulations that form part of RCEMIP (the Radiative‐Convective Equilibrium Model Intercomparison Project). Analysis shows that the presence of anvil cloud, which tends to enhance aggregation when collocated with anomalously moist environments, is reduced in nearly all models when SSTs are increased, leading to a corresponding reduction in the aggregating influence of cloud‐longwave interactions. We also find that cloud‐longwave radiation interactions are stronger in the majority of General Circulation Models (GCMs), typically resulting in faster aggregation compared to Cloud‐system Resolving Models (CRMs). GCMs that have stronger cloud‐longwave interactions tend to aggregate faster, whereas the influence of circulations is the main factor affecting the aggregation rate in CRMs. Plain Language Summary: The spatial organization of tropical rainstorms has major effects on weather and climate. This organization influences the duration and intensity of these convective storms, and alters the amount of radiation absorbed and emitted by the atmosphere. There is great uncertainty in the response of organization to a warming climate, and this results in one of the largest sources of uncertainty in climate predictions. Climate projections rely on either General Circulation Models (GCMs) that can represent the large‐scale motions, or smaller high‐resolution models that represent small‐scale features like cloud formations, but not the large motions. In this study, we compare convective organization in GCMs and Cloud‐system Resolving Models (CRMs) across a range of sea surface temperatures (SSTs). We find that the cloud‐radiation feedbacks that make the convective environment more favorable for further convection, and the non‐convective environment less favorable for convection, are stronger in GCMs than CRMs on average. This is related to larger cloud amounts in GCMs, leading GCMs to have typically faster organization than CRMs. We find these feedbacks which drive aggregation decrease as SST increases, yet the aggregation rate is largely insensitive to SST because of the decrease in the effect of atmospheric motions that oppose aggregation. Key Points: General Circulation Models (GCMs) aggregate faster than Cloud‐system Resolving Models (CRMs) on average due to an enhanced longwave feedbackFeedbacks tend to decrease in magnitude as sea surface temperature increases, although the rate of aggregation remains similarAggregation rate in GCMs is correlated with diabatic feedbacks, while in CRMs it is more related to advection feedbacks [ABSTRACT FROM AUTHOR]

The potential scope of superparameterization (SP) was extended to higher resolutions of the global climate model (GCM) component by devising a technique called blockwise coupling. In this method, a horizontal average of multiple GCM columns, instead of one, is coupled to a cloud‐resolving model (CRM) domain. This enables SP‐GCMs to reduce the computational cost drastically, enabling higher‐resolution GCMs to be superparameterized. A blockwise‐coupled SP‐GCM called SP‐MIROC was implemented by coupling the climate model MIROC6 to the CRM SCALE‐RM. The 4 × 4‐bundled SP‐MIROC successfully reproduced horizontal patterns and frequency distributions of precipitation and realistic amplitudes of equatorial Kelvin waves (EKWs), which were underestimated in the standard MIROC6. As discussed in Yamazaki and Miura (2024b, https://doi.org/10.1029/2023MS003837) of this study, the amplitude boost of EKWs was enabled by a top‐heavy heating in SP‐MIROC. Comparison of power spectra between the 4 × 4‐bundled SP‐MIROC and nonbundled SP‐MIROC indicated that the effective resolution of dynamic variables was not degraded by the blockwise technique. Rather, spectra in the 4 × 4‐bundled SP‐MIROC were more realistic than those in the nonbundled SP‐MIROC. Although the 4 × 4‐bundling limits convective coupling in the smallest GCM scale, it could offer the best match of resolutions between the GCM‐handled dynamics and SP‐derived physics because the effective resolution of the dynamics is lower than the nominal grid spacing. Plain Language Summary: Superparameterization is a technique to explicitly simulate subgrid clouds in low‐resolution global climate models (GCMs) by coupling small domains of a high‐resolution cloud‐resolving model (CRM). This study expands the superparameterization technique by proposing and verifying blockwise coupling, in which a CRM domain is coupled to an average of multiple GCM columns; in contrast, the existing superparameterized (SP) models employ one‐to‐one coupling between a CRM and a GCM column. We implemented a SP‐GCM by coupling the climate model MIROC6 and the SCALE‐RM CRM with 4 × 4‐to‐one blockwise coupling. Multiple benefits in existing SP‐GCMs are reproduced in the simulation results of the blockwise‐coupled SP‐MIROC. Furthermore, 4 × 4‐bundled SP‐MIROC outperformed the one‐to‐one SP‐MIROC in terms of reproducing realistic zonal spectra. Blockwise coupling will drastically reduce computational costs, making superparameterization in higher‐resolution GCMs affordable. Key Points: A superparameterized version of the climate model MIROC6 named superparameterized (SP)‐MIROC was developedComputational cost is reduced by coupling a cloud‐resolving model domain to a block of multiple global climate model columns, instead of oneBlockwise‐coupled SP‐MIROC reproduced most benefits of SP without losing the effective resolution of dynamic variables [ABSTRACT FROM AUTHOR]

Abstract A set of idealized experiments are developed using the Community Atmosphere Model (CAM) to understand the vertical velocity response to reductions in forcing scale that is known to occur when the horizontal resolution of the model is increased. The test consists of a set of rising bubble experiments, in which the horizontal radius of the bubble and the model grid spacing are simultaneously reduced. The test is performed with moisture, through incorporating moist physics routines of varying complexity, although convection schemes are not considered. Results confirm that the vertical velocity in CAM is to first‐order, proportional to the inverse of the horizontal forcing scale, which is consistent with a scale analysis of the dry equations of motion. In contrast, experiments in which the coupling time step between the moist physics routines and the dynamical core (i.e., the “physics” time step) are relaxed back to more conventional values results in severely damped vertical motion at high resolution, degrading the scaling. A set of aqua‐planet simulations using different physics time steps are found to be consistent with the results of the idealized experiments.

Advances in high‐performance computing have enabled large‐eddy simulations (LES) of turbulence, convection, and clouds. However, their potential to improve parameterizations in global climate models (GCMs) is only beginning to be harnessed, with relatively few canonical LES available so far. The purpose of this paper is to begin creating a public LES library that expands the training data available for calibrating and evaluating GCM parameterizations. To do so, we use an experimental setup in which LES are driven by large‐scale forcings from GCMs, which in principle can be used at any location, any time of year, and in any climate state. We use this setup to create a library of LES of clouds across the tropics and subtropics, in the present and in a warmer climate, with a focus on the transition from stratocumulus to shallow cumulus over the East Pacific. The LES results are relatively insensitive to the choice of host GCM driving the LES. Driven with large‐scale forcing under global warming, the LES simulate a positive but weak shortwave cloud feedback, adding to the accumulating evidence that low clouds amplify global warming. Plain Language Summary: Clouds remain one of the largest uncertainties in our understanding and in predictions of climate change because it is challenging to represent their small‐scale dynamics in climate models. However, high‐resolution simulations can provide faithful simulations of clouds and turbulence in limited areas, and these can be used to calibrate climate models. So far, only a limited set of simulations has been used for calibration of climate models, with focus on a few specific locations. This study presents an experimental setup that allows the high‐resolution simulations to be run anywhere on the globe, in any climate state, driven by output from different climate models. The setup is used to create a library of high‐resolution simulations of clouds across the tropics and subtropics in both the current and a warmer climate. The library substantially expands the data set available for the calibration and evaluation of climate models. Key Points: A library of high‐resolution simulations of clouds is created with large‐eddy simulations (LES) driven by a few global climate models (GCMs)Clouds simulated by LES driven by different GCMs are more similar to one another than the corresponding GCM‐simulated cloudsUnder global warming, LES‐simulated low clouds exhibit a weak but positive feedback on warming [ABSTRACT FROM AUTHOR]

Canonical understanding based on general circulation models (GCMs) is that the atmospheric circulation response to midlatitude sea‐surface temperature (SST) anomalies is weak compared to the larger influence of tropical SST anomalies. However, the ∼100‐km horizontal resolution of modern GCMs is too coarse to resolve strong updrafts within weather fronts, which could provide a pathway for surface anomalies to be communicated aloft. Here, we investigate the large‐scale atmospheric circulation response to idealized Gulf Stream SST anomalies in Community Atmosphere Model (CAM6) simulations with 14‐km regional grid refinement over the North Atlantic, and compare it to the responses in simulations with 28‐km regional refinement and uniform 111‐km resolution. The highest resolution simulations show a large positive response of the wintertime North Atlantic Oscillation (NAO) to positive SST anomalies in the Gulf Stream, a 0.4‐standard‐deviation anomaly in the seasonal‐mean NAO for 2°C SST anomalies. The lower‐resolution simulations show a weaker response with a different spatial structure. The enhanced large‐scale circulation response results from an increase in resolved vertical motions with resolution and an associated increase in the influence of SST anomalies on transient‐eddy heat and momentum fluxes in the free troposphere. In response to positive SST anomalies, these processes lead to a stronger and less variable North Atlantic jet, as is characteristic of positive NAO anomalies. Our results suggest that the atmosphere responds differently to midlatitude SST anomalies in higher‐resolution models and that regional refinement in key regions offers a potential pathway to improve multi‐year regional climate predictions based on midlatitude SSTs. Plain Language Summary: Variations in the ocean surface temperature (SST) influence the atmospheric circulation and thus climate over land. Canonical understanding is that tropical SSTs are more important than SSTs in midlatitudes. However, this understanding is based on climate models that don't resolve processes at scales less than 100 km. Here, we show that by increasing the atmospheric model resolution to resolve features on smaller scales, such as weather fronts, we find a larger atmospheric circulation response to midlatitude SST anomalies in the North Atlantic. North Atlantic SST anomalies can be predicted multiple years in advance, and a larger atmospheric circulation response to these predictable SST anomalies therefore implies increased predictability of climate over the surrounding land regions. Key Points: There is a large circulation response to idealized Gulf Stream sea‐surface temperature (SST) anomalies in an atmospheric model with 14‐km regional grid refinementThis response is weaker or absent in simulations with 28‐km or coarser resolution, which do not fully resolve mesoscale frontal processesTransient‐eddy fluxes of heat and momentum are modified as fronts pass over warm SSTs, leading to a large‐scale circulation response [ABSTRACT FROM AUTHOR]

While low‐resolution climate models at present struggle to appropriately simulate convectively coupled large‐scale atmospheric disturbances such as equatorial Kelvin waves (EKWs), superparameterization helps better reproduce such phenomena. To evaluate such model differences based on physical mechanisms, a linearized theoretical framework of convectively coupled EKWs was developed in a form readily applicable to model evaluation by allowing background stability and diabatic heating to have arbitrary vertical profiles rather than assuming simplified ones. A system of linearized equations of convection‐coupled gravity waves was derived as a two‐dimensional model of the convectively coupled EKWs. In this work, the basic states are taken from observations, CTL‐MIROC and SP‐MIROC experiments introduced in Part 1. The tendency of convectively coupled gravity waves to grow faster under top‐heavy heating is confirmed for realistic stratification profiles, as found in previous studies under constant stratifications. A comparison of linear unstable solutions with basic states taken from SP‐MIROC and CTL‐MIROC shows that the top‐heavy heating profile in SP‐MIROC largely contributes to the enhancement of the EKW‐like unstable modes, while subtle differences of stratification profiles considerably affect EKW behaviors. The bottom‐heavy heating bias in the CTL‐MIROC likely originates from insufficient modeling of subgrid stratiform precipitation in tropical organized systems. It is desirable to incorporate such stratiform components in cumulus parameterizations to achieve better EKW reproducibility. Plain Language Summary: Present global climate models (GCMs) cannot resolve small‐scale cumulus convection. Hence, they cannot reproduce sufficient amplitudes of convectively coupled large‐scale atmospheric disturbances such as the equatorial Kelvin wave (EKW). In contrast, high‐resolution models that explicitly simulate cumulus convection can reproduce these disturbances better. In this study, a linearized theoretical framework of the convectively coupled EKW was developed to interpret and evaluate model behavior. Using this framework, EKWs simulated by different models were compared based on their physical mechanisms. The basic states were taken from observations and MIROC simulations introduced in Part 1. The tendency of EKWs to grow under a top‐heavy heating condition was confirmed under realistic stratification profiles as found in previous studies under constant stratifications. A comparison of linear unstable solutions with the basic states taken from the MIROC simulations showed that the heating altitude is the most important factor for enhancing the EKW‐like unstable modes, while stratification profiles can sometimes considerably affect EKW growth. Bottom‐heavy heating bias in the conventional MIROC likely originates from insufficient parameterization of organized precipitating systems in the tropics. It is desirable to incorporate such stratiform components in the cumulus parameterization to achieve better EKW reproducibility. Key Points: A linearized model of the EKW is derived in a form readily applicable to evaluation of numerical simulationsSuperparameterized MIROC enhanced EKW amplitudes through top‐heavy heating profilesTo improve EKW reproducibility, GCMs need better representation of tropical stratiform precipitation [ABSTRACT FROM AUTHOR]

This study presents the development of a so‐called Turbulent Kinetic Energy (TKE)‐l, or TKE‐l, parameterization of the diffusion coefficients for the representation of turbulent diffusion in neutral and stable conditions in large‐scale atmospheric models. The parameterization has been carefully designed to be completely tunable in the sense that all adjustable parameters have been clearly identified and the number of parameters has been minimized as much as possible to help the calibration and to thoroughly assess the parametric sensitivity. We choose a mixing length formulation that depends on both static stability and wind shear to cover the different regimes of stable boundary layers. We follow a heuristic approach for expressing the stability functions and turbulent Prandlt number in order to guarantee the versatility of the scheme and its applicability for planetary atmospheres composed of an ideal and perfect gas such as that of Earth and Mars. Particular attention has been paid to the numerical stability and convergence of the TKE equation at large time steps, an essential prerequisite for capturing stable boundary layers in General Circulation Models (GCMs). Tests, parametric sensitivity assessments and preliminary tuning are performed on single‐column idealized simulations of the weakly stable boundary layer. The robustness and versatility of the scheme are assessed through its implementation in the Laboratoire de Météorologie Dynamique Zoom GCM and the Mars Planetary Climate Model and by running simulations of the Antarctic and Martian nocturnal boundary layers. Plain Language Summary: In planetary atmospheres, turbulent motions actively contribute to the mixing of quantities such as heat, momentum, and chemical species. Such motions are not resolved in coarse‐grid atmospheric models and have to be parameterized. The parameterization of turbulent mixing should ideally be based on physical laws and sufficiently sophisticated to realistically represent the full spectrum of motions over the full range of stability encountered in the atmospheres. However, it also necessarily contains a number of closure parameters not always well identified and whose values are determined empirically—thereby questioning the universality of the parameterization and its potential application over the full globe or even to other planets—or adjusted to guarantee the numerical stability of the model. This study presents the design of a turbulent mixing parameterization that can be fully calibrated and applied in planetary atmospheres such as that of Mars. We then calibrate the parameterization on an idealized simulation set‐up and test its robustness and performance by running simulations of the Antarctic and Martian atmospheres. Key Points: A simple TKE‐l turbulent diffusion scheme is developed in a semi‐heuristic way for applications in models of the Earth and Mars atmospheresAll adjustable parameters are clearly identified and the number of parameters is minimized to thoroughly assess the parametric sensitivityOnce tuned on GEWEX Atmospheric Boundary Layer Study 1 1D simulations, the scheme is able to capture the Antarctic and Martian stable boundary layers in 3D simulations [ABSTRACT FROM AUTHOR]

Representing mixing state of black carbon (BC) is challenging for global climate models (GCMs). The Community Atmosphere Model version 6 (CAM6) with the four‐mode version of the Modal Aerosol Module (MAM4) represents aerosols as fully internal mixtures with uniform composition within each aerosol mode, resulting in high degree of internal mixing of BC with non‐BC species and large mass ratio of coating to BC (RBC, the mass ratio of non‐BC species to BC in BC‐containing particles). To improve BC mixing state representation, we coupled a machine learning (ML) model of BC mixing state index trained on particle‐resolved simulations to the CAM6 with MAM4 (MAM4‐ML). In MAM4‐ML, we use RBC to partition accumulation mode particles into two new modes, BC‐free particles and BC‐containing particles. We adjust RBC to make the modeled BC mixing state index (χmode) match the one predicted by the ML model (χML). On a global average, the mass fraction of BC‐containing particles in accumulation mode decreases from 100% (MAM4‐default) to 48% (MAM4‐ML). The globally averaged χmode decreases from 78% (MAM4‐default) to 63% (MAM4‐ML, 19% reduction) and agrees well with χML (66%). The RBC decreases by 52% for accumulation mode and better agrees with observations. The hygroscopicity drops by 9% for BC‐containing particles in accumulation mode, leading to a 20% reduction in the BC activation fraction. The surface BC concentration increases most (6.9%) in the Arctic, and the BC burden increases by 4%, globally. Our study highlights the application of the ML model for improving key aerosol processes in GCMs. Plain Language Summary: Black carbon (BC) mixing state is an important property for assessing impacts of BC on human health and climate. It describes how BC is distributed among the particle population and within individual particles. Due to computational costs, global climate models (GCMs) consider aerosols as either external mixtures (e.g., bulk models) or fully internal mixtures with uniform composition within a certain size range (e.g., an aerosol bin in sectional models or an aerosol mode in modal models). To improve BC mixing state representation, we coupled a machine learning (ML) model learned from particle‐resolved simulations to the online simulations of the Community Atmosphere Model version 6 (CAM6) with the four‐mode version of the Modal Aerosol Module (MAM4‐ML). We used the mass ratio of coating to BC (RBC) to determine the mass fraction of BC‐free particles and BC‐containing particles, and we adjusted RBC to make the modeled BC mixing state match ML model. The MAM4‐ML model reduces the degree of internal mixture of BC substantially. In addition, MAM4‐ML RBC is reduced by 52%, leading to a 20% reduction in BC activation fraction and a 4% increase in BC burden. This study proposes a method for the application of the ML to improve GCMs. Key Points: Machine learning model trained on particle‐resolved simulations has been coupled to CAM6 MAM4 to improve BC mixing state representationThe BC mixing state index is reduced by 19%, and the mass ratio of coating to BC is reduced by 52%, in better agreement with observationsDue to the drop in hygroscopicity of BC‐containing aerosol (9%), BC activation fraction decreased by 20%, and BC burden increased by 4% [ABSTRACT FROM AUTHOR]

When the negatively buoyant air in the cloud downdrafts reaches the surface, it spreads out horizontally, producing cold pools. A cold pool can trigger new convective cells. However, when combined with the ambient vertical wind shear, it can also connect and upscale them into large mesoscale convective systems (MCS). Given the broad spectrum of scales of the atmospheric phenomenon involving the interaction between cold pools and the MCS, a parameterization was designed here. Then, it is coupled with a classical convection parameterization to be applied in an atmospheric model with an insufficient spatial resolution to explicitly resolve convection and the sub‐cloud layer. A new scalar quantity related to the deficit of moist static energy detrained by the downdrafts mass flux is proposed. This quantity is subject to grid‐scale advection, mixing, and a sink term representing dissipation processes. The model is then applied to simulate moist convection development over a large portion of tropical land in the Amazon Basin in a wet and dry‐to‐wet 10‐days period. Our results show that the cold pool edge parameterization improves the organization, longevity, propagation, and severity of simulated MCS over the Amazon and other different continental areas. Plain Language Summary: In nature, cold pools are formed by cold air masses descending from the low to mid‐troposphere in thunderstorms. When these drafts reach the surface, they spread out horizontally. A manifestation of cold pools is the relatively high speed at the gust front, which can lift environmental air producing new convective cells. Moreover, depending on ambient conditions, the cold pools may help organize the new convective cells, increasing their aggregation and forming the so‐called mesoscale convective systems (MCSs). MCSs, which cover hundreds to thousands of km2, significantly impact the global scale circulation, energy budget, hydrological cycle, and population safety. Forecasting MCSs is challenging for global circulation models (GCM) due to the broad spectrum of scales of the involved atmospheric phenomenon. The computational limitations, at present and for some time to come, do not allow running in real‐time GCMs, which explicitly solves all relevant scales of motion. This paper describes a methodology to account for essential interplays between cold pools edges and moist convection to be applied in the GCMs of weather and climate forecasting. We show that the method improves the model simulation of the main types of MCSs over the Amazon Basin and other continental areas. Key Points: A model scheme for including effect of cold pools edges in triggering new convective cells and storm propagation is presentedThe scheme is coupled with a convection parameterization and applied in the modeling of moist convective systemsThe method improves the organization, longevity, propagation, and severity of the simulated mesoscale convective systems [ABSTRACT FROM AUTHOR]

Proxy records that document the last 2000 years of climate provide evidence for the wide range of the natural climate variability from inter‐annual to secular timescales not captured by the short window of recent direct observations. Assessing climate models ability to reproduce such natural variations is crucial to understand climate sensitivity and impacts of future climate change. Paleoclimate data assimilation (PDA) offers a powerful way to extend the short instrumental period by optimally combining the physics described by General Circulation Climate Models (GCMs) with information from available proxy records while taking into account their uncertainties. Here we present a new PDA approach based on a sequential importance resampling (SIR) Particle filter (PF) that uses Linear Inverse Modeling (LIM) as an emulator of several CMIP‐class GCMs. We examine in a perfect‐model framework the skill of the various LIMs to forecast the dynamics of the surface temperatures and provide spatial field reconstructions over the last millennium in a SIR PF. Our results show that the LIMs allow for skillful ensemble forecasts at 1‐year lead‐time based on GCMs dynamical knowledge with best prediction in the tropics and the North Atlantic. The PDA further provides a set of physically consistent spatial fields allowing robust uncertainty quantification related to climate models biases and proxy spatial sampling. Our results indicate that the LIM yields dynamical memory improving climate variability reconstructions and support the use of the LIM as a GCM‐emulator in real reconstruction to propagate large ensembles of particles at low cost in SIR PF. Plain Language Summary: To detect and attribute anthropogenic climate changes, it is necessary to quantify and understand the wide range of the natural climate variability. The length of the instrumental period is relatively short for investigating slow climate features and their regional impacts. The field of paleoclimate data assimilation offers a way to extend the window of observation and provide physically consistent spatial fields by combining information from both CMIP‐class climate models and proxy records documenting the last 2000 years. These approaches are however confronted with limitations due to the use of expensive global climate models over long periods and the very large number of simulations required to correctly describe the space of possible climate states, which increases exponentially with the dimension of the problem. Here we present a new Proxy Data Assimilation strategy that relies on a statistical model called LIM to reproduce the spatiotemporal dynamics of the surface temperatures simulated by costly CMIP‐class climate models. This climate model emulator is then used to simulate large sets of particles that are optimally combined with the information provided by proxy records to deduce skillful spatial climate fields reconstructions and their low‐frequency variability over the Common Era. Key Points: Linear inverse modeling trained as emulators of CMIP‐class models produce realistic climate and skillful ensemble forecasts as compared to the host‐based modelAn online particle filter incorporating LIMs (SIR‐LIM) overcomes the so‐called data assimilation "curse of dimensionality" problemSIR‐LIM reconstructions for the last millennium allow for uncertainty quantifications due to climate model biases and proxy spatial sampling [ABSTRACT FROM AUTHOR]

Ocean dynamic sea level (DSL) change is a key driver of relative sea level (RSL) change. Projections of DSL change are generally obtained from simulations using atmosphere‐ocean general circulation models (GCMs). Here, we develop a two‐layer climate emulator to interpolate between emission scenarios simulated with GCMs and extend projections beyond the time horizon of available simulations. This emulator captures the evolution of DSL changes in corresponding GCMs, especially over middle and low latitudes. Compared with an emulator using univariate pattern scaling, the two‐layer emulator more accurately reflects GCM behavior and captures non‐linearities and non‐stationarity in the relationship between DSL and global‐mean warming, with a reduction in global‐averaged error during 2271–2290 of 36%, 24%, and 34% in RCP2.6, RCP4.5, and RCP8.5, respectively. Using the emulator, we develop a probabilistic ensemble of DSL projections through 2300 for four scenarios: Representative Concentration Pathway (RCP) 2.6, RCP 4.5, RCP 8.5, and Shared Socioeconomic Pathway (SSP) 3–7.0. The magnitude and uncertainty of projected DSL changes decrease from the high‐to the low‐emission scenarios, indicating a reduced DSL rise hazard in low‐ and moderate‐emission scenarios (RCP2.6 and RCP4.5) compared to the high‐emission scenarios (SSP3‐7.0 and RCP8.5). Plain Language Summary: As climate warms, sea‐level rise poses a major threat to coastal communities and ecosystems. A key driver of local sea level change is ocean dynamic sea level change, which is associated with changes in ocean density and circulation. The primary tools used to project changes in dynamic sea level are atmosphere‐ocean general circulation models, which are computationally expensive. Here we develop an emulator for dynamic sea level changes that is, built upon a two‐layer energy balance model. Considering both the rapid response of well‐mixed surface layer and the delayed response of the deep ocean to forcing, the emulator facilitates the estimation of probability distributions of future dynamic sea level change under different emission scenarios and the extension of projections beyond the time horizon of available simulations from some atmosphere‐ocean general circulation models. This emulator captures the evolution of dynamic sea level changes in the atmosphere‐ocean general circulation models to which it is tuned, including the non‐linear and non‐stationary relationship between the dynamic sea level and global warming. The emulator thus facilitates the estimation of future local sea‐level changes. Key Points: An emulator for DSL changes is developed based on a two‐layer energy balance model and a two‐layer pattern scaling techniqueThe two‐layer emulator can better capture the evolution of DSL in corresponding coupled GCMs than scaling of global mean surface temperatureThe two‐layer emulator allows estimation of the probability of future DSL changes in various emission scenarios over multiple centuries [ABSTRACT FROM AUTHOR]

The path toward realizing the potential of seasonal forecasting and its socioeconomic benefits relies on improving general circulation model (GCM) based dynamical forecast systems. To improve dynamical seasonal forecasts, it is crucial to set up forecast benchmarks, and clarify forecast limitations posed by model initialization errors, formulation deficiencies, and internal climate variability. With huge costs in generating large forecast ensembles, and limited observations for forecast verification, the seasonal forecast benchmarking and diagnosing task proves challenging. Here, we develop a probabilistic deep learning‐based statistical forecast methodology, drawing on a wealth of climate simulations to enhance seasonal forecast capability and forecast diagnosis. By explicitly modeling the internal climate variability and GCM formulation differences, the proposed Conditional Generative Forecasting (CGF) methodology enables bypassing crucial barriers in dynamical forecast, and offers a top‐down viewpoint to examine how complicated GCMs encode the seasonal predictability information. We apply the CGF methodology for global seasonal forecast of precipitation and 2 m air temperature, based on a unique data set consisting 52,201 years of climate simulation. Results show that the CGF methodology can faithfully represent the seasonal predictability information encoded in GCMs. We successfully apply this learned relationship in real‐world seasonal forecast, achieving competitive performance compared to dynamical forecasts. Using this CGF as benchmark, we reveal the impact of insufficient forecast spread sampling that limits the skill of the considered dynamical forecast system. Finally, we introduce different strategies for composing ensembles using the CGF methodology, highlighting the potential for leveraging the strengths of multiple GCMs to achieve advantgeous seasonal forecast. Plain Language Summary: Seasonal forecast benefits a broad range of societal sectors. However, current dynamical seasonal forecast systems are considerably hindered by observation, model, and computation limitations. We develop a machine learning probabilistic forecast model that learns from climate simulations to infer possible climate patterns a season ahead. The model achieves competitive performance for global precipitation and temperature forecast, compared to the costly dynamical forecasts. More importantly, it offers crucial implications for understanding the limitations and improving the current dynamical forecast systems. Key Points: We develop the Conditional Generative Forecasting (CGF) methodology that learns from climate simulation for probabilistic seasonal forecastWe create a unique data set consisting 52,201 years of climate simulation to support data‐driven seasonal forecastsWe demonstrate the competitive skill of the CGF methodology for global seasonal forecast of precipitation and 2 m air temperature [ABSTRACT FROM AUTHOR]

Accurately representing mixed‐phase clouds (MPCs) in global climate models (GCMs) is critical for capturing climate sensitivity and Arctic amplification. Secondary ice production (SIP), can significantly increase ice crystal number concentration (ICNC) in MPCs, affecting cloud properties and processes. Here, we introduce a machine‐learning (ML) approach, called Random Forest SIP (RaFSIP), to parameterize SIP in stratiform MPCs. RaFSIP is trained on 16 grid points with 10‐km horizontal spacing derived from a 2‐year simulation with the Weather Research and Forecasting (WRF) model, including explicit SIP microphysics. Designed for a temperature range of 0 to −25°C, RaFSIP simplifies the description of rime splintering, ice‐ice collisional break‐up, and droplet‐shattering using only a limited set of inputs. RaFSIP was evaluated offline before being integrated into WRF, demonstrating its stable online performance in a 1‐year simulation keeping the same model setup as during training. Even when coupled with the 50‐km grid spacing domain of WRF, RaFSIP reproduces ICNC predictions within a factor of 3 when compared to simulations with explicit SIP microphysics. The coupled WRF‐RaFSIP scheme replicates regions of enhanced SIP and accurately maps ICNCs and liquid water content, particularly at temperatures above −10°C. Uncertainties in RaFSIP minimally impact surface cloud radiative forcing in the Arctic, resulting in radiative biases under 3 Wm−2 compared to simulations with detailed microphysics. Although the performance of RaFSIP in convective clouds remains untested, its adaptable nature allows for data set augmentation to address this aspect. This framework opens possibilities for GCM simplification and process description through physics‐guided ML algorithms. Plain Language Summary: Being able to correctly simulate the amount of ice and liquid in clouds is essential for accurate predictions of the cloud radiative forcing in the climatologically sensitive polar regions. A number of collisional processes between ice and liquid particles in clouds, known as secondary ice production, can significantly enhance the ice crystal number concentrations contained in them. This enhancement is often accompanied by a decrease in the cloud liquid water content, resulting in less opaque clouds to incoming solar radiation, which, in turn, can cause a cloud‐induced warming at the surface. Currently most global climate models are missing the description of the most important secondary ice production processes, which can lead to a biased radiative impact of clouds at the surface. To address this, we propose using a machine learning algorithm trained on high‐resolution model outputs to include the effect of ice multiplication in large‐scale climate models. The machine learning framework effectively captures the physical processes underlying secondary ice production in stratiform clouds using only a few inputs readily available in model frameworks. This approach has the potential to improve model predictions bringing them closer to the observed cloud phase partitioning. Key Points: A random‐forest parameterization for secondary ice production is developed using outputs from a 10‐km horizontal grid spacing simulationCloud phase partitioning agrees within a factor of 3, with radiative biases below 3 Wm−2 compared to the detailed microphysics simulationThe scheme can be adjusted to coarser resolutions typical of climate models without losing computational efficiency and numerical stability [ABSTRACT FROM AUTHOR]

Results are presented from a series of sensitivity tests in idealized global warming experiments using the global nonhydrostatic model, NICAM, in which convection at scales of 7-14 km is explicitly resolved. All have a strong positive longwave cloud feedback larger than that seen in conventional GCMs with parameterized convection. Consequently, the global mean net outgoing radiation decreases in response to increased sea surface temperatures. Large increases in high clouds with tops between 180 and 50 hPa are found, and these changes contribute the most to this longwave cloud feedback. Relative humidity and upper tropospheric temperature also increases strongly, again more so than typically seen in conventional GCMs. The magnitude of the response varies considerably between different versions of NICAM. Most of the NICAM control simulations show large overestimates in cloud fraction between 180 and 50 hPa compared to observations. The changes in cloud fraction in the upper troposphere are strongly correlated with their control values. Versions of NICAM with stronger cloud feedbacks have large positive biases in high-top cloud amount and temperature in the free troposphere in their control simulations. The version which has the best agreement with the observations in this regard has the weakest longwave cloud feedback; however, this is still more strongly positive than that typically seen in conventional GCMs. These results demonstrate the potential for stronger high cloud fraction feedbacks in climate warming scenarios than currently predicted by conventional GCMs and highlight the potential relevance of deep convective processes. [ABSTRACT FROM AUTHOR]

Atmospheric black carbon (BC) aerosols have been long‐lasting uncertain components in environmental and climate studies. Global climate models (GCMs) potentially overestimate BC absorption efficiency due to a lack of consideration of complex BC microphysical and mixing properties. We extract multiple BC properties from observations and develop an aerosol optical module known as Advanced Black Carbon (ABC) in the framework of the Modal Aerosol Model version 4 (MAM4). The ABC module is implemented in the Community Atmosphere Model version 6 (CAM6) and evaluated by in situ and remote sensing observations. CAM6‐ABC addresses the shortcomings of CAM6‐MAM4 in terms of BC microphysical and mixing properties, particularly their size, mixing state and optical simulations. Sensitivity simulations show that the global BC absorption aerosol optical depth at 550 nm simulated by CAM6‐ABC is reduced by ∼29% compared with that in CAM6‐MAM4. The BC absorption enhancement simulated by CAM6‐ABC is reduced from ∼2.6 of the default MAM4 to ∼1.4, which is closer to the observed values (mostly less than 1.5). With improved BC absorption estimation, the biases of aerosol single‐scattering coalbedo simulations are reduced by 18%–69% compared with global Aerosol Robotic Network observations. Moreover, the globally averaged BC direct radiative effect is reduced from 0.37 to 0.28 W/m2 at the top of the atmosphere. Our new scheme alleviates the overestimation of BC absorption in GCMs by constraining BC microphysical and mixing properties when assessing aerosol radiative and climate effects, and it can be easily implemented in most modal‐based aerosol modules of climate models. Plain Language Summary: As a climate warming agent in the atmosphere, black carbon (BC) aerosols remain largely uncertain in their light absorption. In global climate models (GCMs), the BC absorption efficiency tends to be overestimated due to insufficient consideration of the BC microphysical properties and mixing state. Our work quantifies the aerosol representation deficiency in GCMs due to the microphysical and optical properties and compensates for this deficiency in our improved BC and aerosol optical property representation. In this study, we analyze and summarize the key BC properties that highly influence their radiative effects based on multisource observations, for example, size distribution and mixing state, develop an improved aerosol optical module to better account for those observations, and implement this representation in a GCM. We use in situ observations to evaluate our module with improved BC and aerosol optical property representations. The results show that the overestimated BC absorption is much closer to the observations due to a more accurate representation of BC‐related microphysical and mixing properties. Correspondingly, our improved BC representation weakens the modeled BC radiative and climate effects by as much as ∼24%. Key Points: Observations of black carbon (BC) properties are implemented in the current aerosol optical module of Community Atmosphere Model version 6The modeled BC absorption enhancement and aerosol bulk optical properties are closer to the observationsThe BC top of the atmosphere direct radiative effect decreases from 0.37 to 0.28 W/m2 [ABSTRACT FROM AUTHOR]

With the recent advances in data science, machine learning has been increasingly applied to convection and cloud parameterizations in global climate models (GCMs). This study extends the work of Han et al. (2020, https://doi.org/10.1029/2020MS002076) and uses an ensemble of 32‐layer deep convolutional residual neural networks, referred to as ResCu‐en, to emulate convection and cloud processes simulated by a superparameterized GCM, SPCAM. ResCu‐en predicts GCM grid‐scale temperature and moisture tendencies, and cloud liquid and ice water contents from moist physics processes. The surface rainfall is derived from the column‐integrated moisture tendency. The prediction uncertainty inherent in deep learning algorithms in emulating the moist physics is reduced by ensemble averaging. Results in 1‐year independent offline validation show that ResCu‐en has high prediction accuracy for all output variables, both in the current climate and in a warmer climate with +4K sea surface temperature. The analysis of different neural net configurations shows that the success to generalize in a warmer climate is attributed to convective memory and the 1‐dimensional convolution layers incorporated into ResCu‐en. We further implement a member of ResCu‐en into CAM5 with real world geography and run the neural‐network‐enabled CAM5 (NCAM) for 5 years without encountering any numerical integration instability. The simulation generally captures the global distribution of the mean precipitation, with a better simulation of precipitation intensity and diurnal cycle. However, there are large biases in temperature and moisture in high latitudes. These results highlight the importance of convective memory and demonstrate the potential for machine learning to enhance climate modeling. Plain Language Summary: The representation of storms and clouds through empirical algorithms known as parameterizations in global climate models (GCMs) is one of the main sources of biases in the simulation of rainfall and atmospheric circulation. Here an ensemble of 8 deep neural networks are used to replace the conventional parameterization of atmospheric moist physics processes. They are trained on data sampled from 1‐year present‐day climate simulation by a "superparameterized" climate model, which uses a two‐dimensional cloud‐scale model to explicitly simulate convection and clouds inside each GCM grid box. On ensemble averaging, the neural nets produce highly accurate predictions of precipitation characteristics including global distribution and intensity. Furthermore, the machine‐learned emulator trained on data in the current climate also represents convection and precipitation extremely well in a warmer climate. A member of the ensemble of the neural nets is implemented into a GCM. The model is then integrated for 5 years, producing reasonable results. Key Points: An ensemble of deep convolutional residual neural networks is used to reduce the uncertainty in moist physics emulationsThe ensemble of the neural networks trained on data from a present‐day climate simulation generalizes well to a +4K warm climate offlineA multi‐year stable online integration is achieved in a real‐geography GCM with reasonable results [ABSTRACT FROM AUTHOR]

It is very well known that parameter tuning can have a major impact on the performance of Global Climate Models. However, parameter tuning is not the only implementation detail that can drastically affect model performance and the representation of various phenomena in models. "Minor‐looking treatments" often exert a critical control on model performance; they include lower and upper limits of physical variables, thresholds of variables that control the enabling or disabling of a specific process, whether two schemes can work together or only one scheme works exclusively, and numerical methods including the order of calling various physics schemes. The impacts of such minor‐looking treatments are sometimes comparable to or even larger than those obtained by introducing advanced parameterizations based on theory and observation. Because the importance of these treatments is often overlooked and not discussed in the literature, we comprehensively summarize examples of various minor‐looking treatments that can considerably affect model performance. Plain Language Summary: Global climate models (GCMs) are used for climate simulations, including global warming simulations. Global climate models (GCMs) simulate various physical processes such as convection, clouds, radiation, and turbulence. A lot of effort is devoted to improving the representation of these physical processes by introducing more sophisticated or more complicated and detailed processes. The values of parameters in such processes significantly affect the model performance, and many modelers carefully set such parameter values. However, other factors can also affect model performance. The lower or upper limits specified in the detailed implementation of such processes, threshold values for switching on a certain effect, how to calculate each effect, vertical resolutions, and so on can significantly affect the model performance. We call these "minor‐looking treatments" and show a wide range of examples in this paper. Key Points: Minor‐looking treatments crucially determine the performance of global climate modelsThey include lower limits, upper limits, thresholds for enabling physical processes, and numerical methodsMinor‐looking treatments should be documented in detail, shared, and discussed more in the climate community [ABSTRACT FROM AUTHOR]

Contemporary general circulation models (GCMs) and Earth system models (ESMs) are developed by a large number of modeling groups globally. They use a wide range of representations of physical processes, allowing for structural (code) uncertainty to be partially quantified with multi‐model ensembles (MMEs). Many models in the MMEs of the Coupled Model Intercomparison Project (CMIP) have a common development history due to sharing of code and schemes. This makes their projections statistically dependent and introduces biases in MME statistics. Previous research has focused on model output and code dependence, and model code genealogy of CMIP models has not been fully analyzed. We present a full reconstruction of CMIP3, CMIP5, and CMIP6 code genealogy of 167 atmospheric models, GCMs, and ESMs (of which 114 participated in CMIP) based on the available literature, with a focus on the atmospheric component and atmospheric physics. We identify 12 main model families. We propose family and ancestry weighting methods designed to reduce the effect of model structural dependence in MMEs. We analyze weighted effective climate sensitivity (ECS), climate feedbacks, forcing, and global mean near‐surface air temperature, and how they differ by model family. Models in the same family often have similar climate properties. We show that weighting can partially reconcile differences in ECS and cloud feedbacks between CMIP5 and CMIP6. The results can help in understanding structural dependence between CMIP models, and the proposed ancestry and family weighting methods can be used in MME assessments to ameliorate model structural sampling biases. Plain Language Summary: Contemporary global climate models are developed by a large number of modeling groups internationally. Commonly, projections from multiple models are used together to calculate multi‐model means and quantify uncertainty. Because many of the models share parts of their computer code, algorithms and parametrization schemes, they are not independent. Overrepresented models can cause biases in multi‐model means, and uncertainty may be underestimated if model dependence is not taken into account. We document a full code genealogy of 167 models, of which 114 participated in the Coupled Model Intercomparison Project (CMIP) phases 3, 5, and 6, with a focus on the atmospheric component. We identify 12 main model families. We show that models in the same family often have similar estimates of key climate properties. We propose statistical weighting methods based on the model family and code relationship, and show that they can reconcile some of the difference in results between the two most recent CMIP phases. The weighting methods or a selection of independent models based on the genealogy can be used in model assessment studies to reduce the effects of model dependence. Key Points: We reconstruct a code genealogy of 167 climate models with a focus on the atmospheric component and atmospheric physicsAll models originate from 12 main model families, and models in the same family often have similar climate feedbacks and sensitivityProposed ancestry and family weighting can partly reconcile differences in means between the Coupled Model Intercomparison Project phases [ABSTRACT FROM AUTHOR]

Regional climate models (RCMs) have a high computational cost due to their higher spatial resolution compared to global climate models (GCMs). Therefore, various downscaling approaches have been developed as a surrogate for the dynamical downscaling of GCMs. This study assesses the potential of using a cost‐efficient machine learning alternative to dynamical downscaling by using the example case study of emulating surface mass balance (SMB) over the Antarctic Peninsula. More specifically, we determine the impact of the training framework by comparing two training scenarios: (a) a perfect and (b) an imperfect model framework. In the perfect model framework, the RCM‐emulator learns only the downscaling function; therefore, it was trained with upscaled RCM (UPRCM) features at GCM resolution. This emulator accurately reproduced SMB when evaluated on UPRCM, but its predictions on GCM data conserved RCM‐GCM inconsistencies and led to underestimation. In the imperfect model framework, the RCM‐emulator was trained with GCM features and downscaled the GCM while exposed to RCM‐GCM inconsistencies. This emulator predicted SMB close to the truth, showing it learned the underlying inconsistencies and dynamics. Our results suggest that a deep learning RCM‐emulator can learn the proper GCM to RCM downscaling function while working directly with GCM data. Furthermore, the RCM‐emulator presents a significant computational gain compared to an RCM simulation. We conclude that machine learning emulators can be applied to produce fast and fine‐scaled predictions of RCM simulations from GCM data. Plain Language Summary: Over the last century, climate scientists have tried to deepen their understanding of the behavior of climate processes through two types of computer climate simulations: global (GCMs) and regional (RCMs) climate models. GCMs cover the whole planet but do not contain fine spatial details, whereas RCMs provide highly detailed information but cover small areas and come at a high additional computational cost. Therefore, we imitated regional models from global models using machine learning to facilitate their faster development. To test our machine learning framework, we focused on the Antarctic Peninsula and aimed to reproduce the surface mass balance (SMB) of ice formation and loss. We trained our model to learn the relationship between a group of low‐resolution images of climate variables and a high‐resolution image from SMB images in the same region. Our results show that the machine learning model is fast and could recreate regional images of ice sheet processes from global data almost identical to existing on‐site observations. This is a good start for further usage of machine learning emulators. In conclusion, we can make fast and detailed reproductions of SMB processes at regional scales from globally accessible climate data using machine learning. Key Points: We developed a computationally fast machine learning emulator to downscale a global climate model (GCM) to regional resolutionThe emulator reproduces regional high‐resolution surface mass balance predictions over the Antarctic Peninsula from a GCMThe imperfect model framework outperforms the perfect model framework in the application success of the deep learning emulator [ABSTRACT FROM AUTHOR]

Climate models struggle to accurately represent the highly reflective boundary layer clouds overlying the remote and stormy Southern Ocean. We use in situ aircraft observations from the Southern Ocean Clouds, Radiation and Aerosol Transport Experimental Study (SOCRATES) to evaluate Southern Ocean clouds in a cloud‐resolving large‐eddy simulation (LES) and two coarse resolution global atmospheric models, the CESM Community Atmosphere Model (CAM6) and the GFDL Atmosphere Model (AM4), run in a nudged hindcast framework. We develop six case studies from SOCRATES data which span the range of observed cloud and boundary layer properties. For each case, the LES is run once forced purely using reanalysis data (fifth generation European Centre for Medium‐Range Weather Forecasts atmospheric reanalysis, "ERA5 based") and once strongly nudged to an aircraft profile("Obs based"). The ERA5‐based LES can be compared with the global models, which are also nudged to reanalysis data and are better for simulating cumulus. The Obs‐based LES closely matches an observed cloud profile and is useful for microphysical comparisons and sensitivity tests and simulating multilayer stratiform clouds. We use two‐moment Morrison microphysics in the LES and find that it simulates too few frozen particles in clouds occurring within the Hallett‐Mossop temperature range. We tweak the Hallett‐Mossop parameterization so that it activates within boundary layer clouds, and we achieve better agreement between observed and simulated microphysics. The nudged global climate models (GCMs) simulate liquid‐dominated mixed‐phase clouds in the stratiform cases but excessively glaciate cumulus clouds. Both GCMs struggle to represent two‐layer clouds, and CAM6 has low droplet concentrations in all cases and underpredicts stratiform cloud‐driven turbulence. Plain Language Summary: The Southern Ocean, the wide band of water North of Antarctica, is the stormiest place on Earth. Weather systems constantly whirl the atmosphere and blanket the ocean in clouds. Low‐lying clouds reflect sunlight back to space and cool the Earth. Here, we investigate how well the computer models that we use to understand the climate and to forecast future climates can simulate these clouds. We use recent aircraft measurements from the Southern Ocean Clouds, Radiation and Aerosol Transport Experimental Study (SOCRATES) to evaluate two leading U.S. global climate models, the GFDL Atmosphere Model (AM4) and the CESM Community Atmosphere Model (CAM6). We additionally run detailed simulations of Southern Ocean clouds over a small area to understand which physical processes are relevant to cloud formation. We find that our detailed simulations include most of the physics that is relevant to low‐lying Southern Ocean clouds, but one particular type of ice multiplication process, called Hallett‐Mossop rime splintering, is not active enough. CAM6 and AM4 make too much ice in, or glaciate, cumulus clouds. CAM6 has too few cloud droplets, and we hypothesize that this is caused by glaciation and by the simulated clouds driving too little turbulent mixing of the atmosphere. Key Points: SAM LES represents diverse Southern Ocean boundary layer and cloud structures wellCAM6 and AM4 maintain supercooled water in stratiform clouds but excessively glaciate cumuliCAM6 underpredicts stratiform cloud‐driven turbulence and cloud droplet concentration [ABSTRACT FROM AUTHOR]

Convections still need to be parameterized in general circulation models (GCMs) in the coming decades. Performances of GCMs are significantly influenced by the convection schemes used. In contrast to most conventional cloud models that ignore in‐cloud inhomogeneities, a new convective cloud model explicitly considering in‐cloud inhomogeneities is developed. The new model adopts a single bulk plume approach, but divides the plume into a series of interacting sub‐plumes in order to mimic the transition from the plume core to its edges. We implemented the new cloud model in NCAR Community Atmosphere Model version 5.0 (CAM5) and evaluated its performance. Single‐column tests show a stronger mass flux profile and low‐level detrainment with the new model. In global simulations, the long‐lasting wet bias in the tropical free troposphere of CAM5 is alleviated. Spatial pattern and intensity distribution of precipitation is improved, but the global hydrological cycle is over‐enhanced. The diurnal cycle of tropical precipitation is also better captured though the precipitation peak still occurs earlier than the observations and the amplitude of the cycle tends to weaken. MJO simulation is slightly improved with the new scheme. In addition, the new cloud model generated more tropical cyclones, in better agreement with observations. Plain Language Summary: Convection scheme significantly impacts the performance of general circulation models. In nature, physical properties inside convective clouds are highly inhomogeneous. For example, convective cloud top reaches higher at its core but lower at its edges. However, most convection schemes use a bulk homogeneous plume to describe convective clouds. Here, we developed a new cloud model that explicitly considered the in‐cloud inhomogeneity by dividing the plume into a series of sub‐plumes in order to mimic the cloud property transition from the core to its edges. Implementation and test of this new cloud model in NCAR CAM5 are introduced. Compared to the default scheme, a stronger mass flux profile and low‐level detrainment are produced in single column simulations. Tropical precipitation spatial distribution, diurnal cycle, and intensity distribution are improved. More tropical cyclones are generated using the new cloud model, in better agreement with observations. Key Points: A new deep convective cloud model considering in‐cloud inhomogeneity is proposedThe new cloud model is implemented and tested in NCAR CAM5Tropical precipitation distribution and diurnal cycle are improved using the new cloud model [ABSTRACT FROM AUTHOR]

Cloud‐phase partitioning has been studied in the context of cloud feedback and climate sensitivity; however, precipitation‐phase partitioning also has a significant role in controlling the energy budget and sea ice extent. Although some global models have introduced a more sophisticated precipitation parameterization to reproduce realistic cloud and precipitation processes, the effects on the process representation of mixed‐ and ice‐phase precipitation are poorly understood. Here, we evaluate how different precipitation modeling (i.e., diagnostic [DIAG] vs. prognostic [PROG] schemes) affects the simulated precipitation phase and occurrence frequency. Two versions of MIROC6 were used with the satellite simulator COSP2. Although the PROG scheme significantly improves the simulated cloud amount and snowfall rates, the phase partitioning, frequency, and intensity of precipitation with the PROG scheme are still biased, and are even worse than with the DIAG scheme. We found a "too frequent and too light" Arctic snowfall bias in the PROG, which cannot be eliminated by model tuning. The cloud‐phase partitioning is also affected by the different approaches used to consider precipitation. The ratio of supercooled liquid water is underrepresented by switching from the DIAG to PROG scheme, because some snowflakes are regarded to be cloud ice. Given that the PROG precipitation retains more snow in the atmosphere, the underestimation becomes apparent when other models incorporate the PROG scheme. This depends on how much precipitation is within the clouds in the model. Our findings emphasize the importance of correctly reproducing the phase partitioning of cloud and precipitation, which ultimately affects the simulated climate sensitivity. Plain Language Summary: This study examined cloud and precipitation phase partitioning (i.e., the ratio between liquid and ice) in the Arctic using the MIROC6 global climate model (GCM). Despite recent advances in precipitation modeling by GCMs, the associations between the macrostructures (i.e., cloud coverage and precipitation rate) and phase partitioning have been little studied. Prognostic treatment of precipitation, which is a more sophisticated parameterization, yields seasonal and annual cloud cover and snowfall that are in better agreement with satellite observations. However, it tends to generate snowfall too frequently and too lightly, resulting in the misrepresentation of precipitation phase partitioning. In addition, there is a risk of overestimating the ratio of cloud ice to cloud liquid by including prognostic precipitation. The bias is difficult to remove by model tuning alone. If the models misrepresent the precipitation phase partitioning, then the bias will further influence feedback processes in a future warming scenario through the snow‐to‐rain phase change, similar to the cloud phase feedback. Our findings emphasize the importance of conducting process‐oriented model evaluations on a regional scale. Key Points: We examined seasonal model biases in precipitation types of Arctic clouds using a satellite simulatorThere are compensating errors between the cloud amount and snowfall, even for the prognostic precipitation schemeRealistic cloud and precipitation processes and their phase partitioning cannot be achieved by model tuning [ABSTRACT FROM AUTHOR]

Assessing and quantifying the uncertainties in projected future changes of energy and water budgets over land surface are important steps toward improving our confidence in climate change projections. In this study, the contribution of land surface models to the inter-GCM variation of projected future changes in land surface energy and water fluxes are assessed based on output from 19 global climate models (GCMs) and offline Community Land Model version 4 (CLM4) simulations driven by meteorological forcing from the 19 GCMs. Similar offline simulations using CLM4 with its dynamic vegetation submodel are also conducted to investigate how dynamic vegetation feedback, a process that is being added to more earth system models, may amplify or moderate the intermodel variations of projected future changes. Projected changes are quantified as the difference between the 2081-2100 period from the Representative Concentration Pathway 8.5 (RCP8.5) future experiment and the 1981-2000 period from the historical simulation. Under RCP8.5, projected changes in surface water and heat fluxes show a high degree of model dependency across the globe. Although precipitation is very likely to increase in the high latitudes of the Northern Hemisphere, a high degree of model-related uncertainty exists for evapotranspiration, soil water content, and surface runoff, suggesting discrepancy among land surface models (LSMs) in simulating the surface hydrological processes and snow-related processes. Large model-related uncertainties for the surface water budget also exist in the Tropics including southeastern South America and Central Africa. These uncertainties would be reduced in the hypothetical scenario of a single near-perfect land surface model being used across all GCMs, suggesting the potential to reduce uncertainties through the use of more consistent approaches toward land surface model development. Under such a scenario, the most significant reduction is likely to be seen in the Northern Hemisphere high latitudes. Including representation of vegetation dynamics is expected to further amplify the model-related uncertainties in projected future changes in surface water and heat fluxes as well as soil moisture content. This is especially the case in the high latitudes of the Northern Hemisphere (e.g., northwestern North America and central North Asia) where the projected vegetation changes are uncertain and in the Tropics (e.g., the Amazon and Congo Basins) where dense vegetation exists. Findings from this study highlight the importance of improving land surface model parameterizations related to soil and snow processes, as well as the importance of improving the accuracy of dynamic vegetation models. [ABSTRACT FROM AUTHOR]

The low cloud bias in global climate models (GCMs) remains an unsolved problem. Coarse vertical resolution in GCMs has been suggested to be a significant cause of low cloud bias because planetary boundary layer parameterizations cannot resolve sharp temperature and moisture gradients often found at the top of subtropical stratocumulus layers. This work aims to ameliorate the low cloud problem by implementing a new computational method, the Framework for Improvement by Vertical Enhancement (FIVE), into the Energy Exascale Earth System Model (E3SM). Three physics schemes representing microphysics, radiation, and turbulence as well as vertical advection are interfaced to vertically enhanced physics (VEP), which allows for these processes to be computed on a higher vertical resolution grid compared to the rest of the E3SM model. We demonstrate the better representation of subtropical boundary layer clouds with FIVE while limiting additional computational cost from the increased number of levels. When the vertical resolution approaches the large eddy simulation‐like vertical resolution in VEP, the climatological low cloud amount shows a significant increase of more than 30% in the southeastern Pacific Ocean. Using FIVE to improve the representation of low‐level clouds does not come with any negative side effects associated with the simulation of mid‐ and high‐level cloud and precipitation, that can occur when running the full model at higher vertical resolution. Plain Language Summary: Most global climate models (GCMs) underestimate low‐level clouds. Increasing vertical resolution in GCMs is intended to address some of the issues contributing to the problem. In this study, we have implemented a new computational method, known as the Framework for Improvement by Vertical Enhancement (FIVE). FIVE can increase the vertical resolution for select aspects of a GCM, and in this study, we apply FIVE to the Energy Exascale Earth System Model. Our results show that when the vertical resolution approaches 5–10 m, the low cloud amount shows a significant increase of more than 30% in the southeastern Pacific Ocean, while the FIVE method also prevents the simulations from being too computationally expensive. Key Points: A novel computational framework, Framework for Improvement by Vertical Enhancement (FIVE), has been implemented into Energy Exascale Earth System Model (E3SM) and allows select physics to be computed on a higher vertical gridWhen the vertical resolution approaches the large eddy simulation‐like in E3SM‐FIVE, the low cloud shows an increase of more than 30% in the Pacific OceanE3SM‐FIVE is much less computationally expensive compared to E3SM with the same high vertical resolution [ABSTRACT FROM AUTHOR]

Nitrate aerosol plays an important role in affecting regional air quality as well as Earth's climate. However, it is not well represented or even neglected in many global climate models. In this study, we couple the Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) module with the four‐mode version of the Modal Aerosol Module (MAM4) in DOE's Energy Exascale Earth System Model version 2 (E3SMv2) to simulate nitrate aerosol and its radiative effects. We find that nitrate aerosol simulated by E3SMv2‐MAM4‐MOSAIC is sensitive to the treatment of gaseous HNO3 transfer to/from interstitial particles related to accommodation coefficients of HNO3 (αHNO3 ${\alpha }_{{\mathrm{H}\mathrm{N}\mathrm{O}}_{3}}$) on dust and non‐dust particles. We compare three different treatments of HNO3 transfer: (a) a treatment (MTC_SLOW) that uses a low αHNO3 ${\alpha }_{{\mathrm{H}\mathrm{N}\mathrm{O}}_{3}}$ in the mass transfer coefficient (MTC) calculation; (b) a dust‐weighted MTC treatment (MTC_WGT) that uses a high αHNO3 ${\alpha }_{{\mathrm{H}\mathrm{N}\mathrm{O}}_{3}}$ on non‐dust particles; and (c) a dust‐weighted MTC treatment that also splits coarse mode aerosols into the coarse dust and sea salt sub‐modes in MOSAIC (MTC_SPLC). MTC_WGT and MTC_SPLC increase the global annual mean (2005–2014) nitrate burden from 0.096 (MTC_SLOW) to 0.237 and 0.185 Tg N, respectively, mostly in the coarse mode. MTC_WGT and MTC_SPLC also produce stronger nitrate direct radiative forcing (−0.048 and −0.051 W m−2, respectively) and indirect forcing (−0.33 and −0.35 W m−2, respectively) than MTC_SLOW (−0.021 and −0.24 W m−2). MTC_WGT and MTC_SPLC improve nitrate surface concentrations over remote oceans based on limited observations and vertical profiles of nitrate concentrations against aircraft measurements below 400 hPa. Plain Language Summary: Atmospheric aerosols play an important role in the Earth's climate system through their effects on radiation and clouds, and their representation continues to be a major uncertainty in global climate models (GCMs). Nitrate aerosol accounts for a notable fraction of total aerosol mass, but it is crudely represented or even neglected in many modern GCMs. In this study, we implement a comprehensive but computationally efficient aerosol chemistry module in the U.S. DOE Energy Exascale Earth System Model version 2, a state‐of‐the‐science GCM, to simulate nitrate aerosols and quantify their radiative effects. Modeled nitrate concentrations are in good agreement with aircraft observations but have positive biases relative to ground‐based network measurements. We also find that simulated nitrate lifecycle is sensitive to the treatment of gaseous HNO3 transfer to/from interstitial particles related to a parameter characterizing the sticking probability of a gas molecule at the surface of different aerosols such as dust and sea salt particles. Key Points: The Model for Simulating Aerosol Interactions and Chemistry module is implemented in Energy Exascale Earth System Model version 2 with Model for Ozone and Related chemical Tracers gas chemistry to simulate nitrate aerosolsModeled nitrate concentrations are in good agreement with aircraft observations but have positive biases at the surfaceTreatments of HNO3 accommodation coefficients and the mixing state of dust and sea salt particles significantly impact nitrate lifecycle [ABSTRACT FROM AUTHOR]

Representing subgrid variability of cloud properties has always been a challenge in global climate models (GCMs). In many cloud microphysics schemes, the warm rain non‐linear process rates calculated based on grid‐mean cloud properties are usually scaled by an enhancement factor (EF) to account for the effects of subgrid cloud variability. In our study, we find that the EF derived from Cloud Layers Unified by Binormals in Community Atmosphere Model version 6 (CAM6) is severely overestimated in most of the cloudy oceanic areas, which leads to strong overestimation of the autoconversion rate. We improve the EF in warm rain simulation by developing a new formula for in‐cloud subgrid cloud water variance. With the updated subgrid cloud water variance and EF treatment, the liquid cloud fraction (LCF) and cloud optical thickness (COT) increases noticeably for marine stratocumulus, and the shortwave cloud forcing (SWCF) matches better with observations. The updated formula improves the relationship between autoconversion rate and cloud droplet number concentration, and it decreases the sensitivity of autoconversion rate to aerosols. The sensitivity of liquid water path to aerosols decreases noticeably and is in better agreement with that in MODIS. Although the sensitivity of COT is similar to that in MODIS, CAM6 underestimates the sensitivity of grid‐mean SWCF to aerosols because of the underestimation in the sensitivities of LCF and in‐cloud SWCF. Our results indicate the importance of representing reasonable subgrid cloud variability in the simulation of cloud properties and aerosol‐cloud interaction in GCMs. Plain Language Summary: Restricted by the coarse horizontal resolution, the global climate model needs to scale the warm rain non‐linear process rates with an enhancement factor to account for the effects of subgrid variability of cloud water. However, this variability is severely overestimated in most of the cloudy oceanic areas in Community Atmosphere Model version 6, which leads to a strong overestimation of the precipitation. We develop a new formula for calculating the subgrid variability of cloud water, which greatly improves the magnitude of the variability and contributes to better simulation of cloud properties and shortwave cloud forcing (SWCF) using satellite observation as a benchmark. The sensitivity of cloud liquid water path to aerosols decreases noticeably with the new formula. The model underestimates the sensitivity of SWCF to aerosols due to the underestimation in the sensitivities of liquid cloud fraction and cloud albedo. Our results indicate the importance of representing reasonable subgrid cloud variability in the simulation of cloud properties and aerosol‐cloud interaction in the climate model. Key Points: A new formula for the in‐cloud subgrid cloud water variance is developed to improve the enhancement factor of autoconversion rateThe simulated sensitivity of cloud liquid water path to aerosols decreases obviously with the updated formulaThe simulated sensitivity of grid‐mean shortwave cloud forcing to aerosols is underestimated compared to MODIS observation [ABSTRACT FROM AUTHOR]

Deficiencies in convection trigger functions, used in deep convection parameterizations in General Circulation Models (GCMs), have critical impacts on climate simulations. A novel convection trigger function is developed using the machine learning (ML) classification model XGBoost. The large‐scale environmental information associated with convective events is obtained from the long‐term constrained variational analysis forcing data from the Atmospheric Radiation Measurement (ARM) program at its Southern Great Plains (SGP) and Manaus (MAO) sites representing, respectively, continental mid‐latitude and tropical convection. The ML trigger is separately trained and evaluated per site, and jointly trained and evaluated at both sites as a unified trigger. The performance of the ML trigger is compared with four convective trigger functions commonly used in GCMs: dilute convective available potential energy (CAPE), undilute CAPE, dilute dynamic CAPE (dCAPE), and undilute dCAPE. The ML trigger substantially outperforms the four CAPE‐based triggers in terms of the F1 score metric, widely used to estimate the performance of ML methods. The site‐specific ML trigger functions can achieve, respectively, 91% and 93% F1 scores at SGP and MAO. The unified trigger also has a 91% F1 score, with virtually no degradation from the site‐specific training, suggesting the potential of a global ML trigger function. The ML trigger alleviates a GCM deficiency regarding the overprediction of convection occurrence, offering a promising improvement to the simulation of the diurnal cycle of precipitation. Furthermore, to overcome the black box issue of the ML methods, insights derived from the ML model are discussed, which may be leveraged to improve traditional CAPE‐based triggers. Plain Language Summary: Deficiencies in convection trigger function, a set of conditions used to determine whether the convection will be activated at a given time in General Circulation Models (GCMs), have critical impacts on model simulated climate. This work presents a novel convection trigger function using a machine learning (ML) model. Environmental information on convective events are obtained from long‐term data from the Atmospheric Radiation Measurement (ARM) program at its Southern Great Plains (SGP) and Manaus (MAO) sites, which represent two distinct convective regimes. The ML trigger is separately trained and evaluated per site, and jointly trained and evaluated at both sites as a unified trigger. The ML trigger substantially outperforms the four CAPE‐based triggers commonly used in GCMs. The unified trigger virtually has no degradation from the site‐specific training, suggesting some promise to develop a global trigger function. The ML trigger alleviates a GCM deficiency regarding the overprediction of convection occurrence, offering a promising improvement to the simulation of the diurnal cycle of precipitation. Furthermore, to overcome the black box issue of the ML methods, insights derived from the ML model are discussed, which may be leveraged to improve traditional CAPE‐based triggers. Key Points: A machine learning convective trigger function greatly outperforms four convective available potential energy (CAPE)‐based triggers at two distinct convective regimesInsights are derived from the machine learning trigger that could be used to improve the existing traditional CAPE‐based triggersResults suggest that a unified machine learning trigger function could be developed for use in climate models [ABSTRACT FROM AUTHOR]

Deep learning (DL) has emerged as a promising tool to downscale climate projections at regional‐to‐local scales from large‐scale atmospheric fields following the perfect‐prognosis approach. Given their complexity, it is crucial to properly evaluate these methods, especially when applied to changing climatic conditions where the ability to extrapolate/generalize is key. In this work, we intercompare several DL models extracted from the literature for the same challenging use‐case (downscaling temperature in the CORDEX North America domain) and expand standard evaluation methods building on eXplainable Artificial Intelligence (XAI) techniques. Specifically, we introduce two novel XAI‐based diagnostics—Aggregated Saliency Map and Saliency Dispersion Maps—and show how they can be used to unravel the internal behavior of these models, aiding in their design and evaluation. This work advocates for the introduction of XAI techniques into deep downscaling evaluation frameworks, especially when working with large regions and/or under climate change conditions. Plain Language Summary: Due to limitations in the computational resources available, General Circulation Models (GCMs) are often used to simulate the climate system over coarse resolution grids. This hampers the applicability of GCM products in the regional‐to‐local scale, highly demanded by different socio‐economic sectors. Statistical downscaling aims to solve this problem by generating high‐resolution climate fields. Recently, machine learning techniques—particularly deep learning (DL) models—have shown promising results in this task. These models are first trained in a historical period through observational data sets, and then applied to the GCM outputs of plausible future scenarios, thus generating high‐resolution climate change products. To assess the performance of these methods, a number of evaluation metrics have been proposed considering both the skill to reproduce present climate conditions and the ability to generalize changing conditions. Here, we illustrate the possibilities of eXplainable Artificial Intelligence (XAI) techniques to expand the evaluation framework for deep downscaling methods, introducing new XAI‐derived diagnostics to unravel their internal behavior. The results show the usefulness of incorporating XAI techniques into statistical downscaling evaluation frameworks, especially when working with large regions and/or under climate change conditions. Key Points: Explainable artificial intelligence (XAI) facilitates the evaluation of deep downscaling models by unraveling their internal behaviorXAI techniques can detect structural problems not revealed by standard evaluation, which may be relevant for understanding model differencesXAI techniques can assess the relevance and locality of the predictors of deep learning models, helping to assess their physical consistency [ABSTRACT FROM AUTHOR]

Precipitation is highly variable on spatial scales smaller than a typical general circulation model (GCM) grid box. Neglecting this subgrid‐scale variability can impact the simulation of the underlying land surface and its coupling with the atmosphere. In this study we present a scheme to stochastically disaggregate precipitation from an atmospheric grid to an underlying mosaic of surface tiles, using an assumed probability distribution of subgrid precipitation derived from observations. Unlike previous GCM‐based schemes, our approach includes a treatment of memory to allow persistence of subgrid features. In single column model experiments, we demonstrate the ability of the scheme to reproduce observed precipitation statistics at the spatial scale of the surface tiles. The root mean square error of hourly spatial standard deviation of precipitation is reduced from 1.90 to 0.96 mm hr−1 when disaggregation is applied. The scheme increases the spatial standard deviations of surface heat fluxes, soil moisture and temperature, and we show that incorporating memory amplifies these increases by a factor of 2–4. We also document increases in mean precipitation runoff and the Bowen ratio. Plain Language Summary: Global climate and weather models (GCMs), used for global warming projections and near‐term weather prediction, represent the atmosphere on a grid, while the underlying land surface is often composed of smaller elements. In most models, all surface elements under a given atmospheric grid point receive the same precipitation rate. However, real‐world precipitation rates vary significantly over the scale of a GCM grid cell. This paper presents an approach to represent subgrid precipitation variability in a GCM. Grid mean precipitation is randomly distributed across surface elements, such that surface precipitation reproduces observed statistics. The approach is found to increase grid cell‐averaged surface runoff as well as the spatial variances of surface temperature, soil moisture, and surface sensible heat and moisture fluxes. Key Points: A parameterization is developed to stochastically distribute precipitation across subgrid surface tiles within a general circulation model (GCM)In single column experiments, the scheme reproduces observed precipitation statistics at the scale of land surface tilesThe scheme increases surface runoff, Bowen ratio, and the spatial variance of surface properties at the ARM Southern Great Plains site [ABSTRACT FROM AUTHOR]

The Walker circulation connects the regions with deep atmospheric convection in the western tropical Pacific to the shallow‐convection, tropospheric subsidence, and stratocumulus cloud decks of the eastern Pacific. The purpose of this study is to better understand the multi‐scale interactions between the Walker circulation, cloud systems, and interactive radiation. To do this we simulate a mock‐Walker Circulation with a full‐physics general circulation model using idealized boundary conditions. Our experiments use a doubly‐periodic domain with grid‐spacing of 1, 2, 25, and 100 km. We thus span the range from General Circulation Models (GCMs) to Cloud‐system Resolving Models (CRMs). Our model is derived from the Geophysical Fluid Dynamics Laboratory atmospheric GCM (AM4.0). We find substantial differences in the mock‐Walker circulation simulated by our GCM‐like and CRM‐like experiments. The CRM‐like experiments have more upper level clouds, stronger overturning circulations, and less precipitation. The GCM‐like experiments have a low‐level cloud fraction that is up to 20% larger. These differences leads to opposite atmospheric responses to changes in the longwave cloud radiative effect (LWCRE). Active LWCRE leads to increased precipitation for our GCMs, but decreased precipitation for our CRMs. The LWCRE leads to a narrower rising branch of the circulation and substantially increases the fraction of precipitation from the large‐scale cloud parameterization. This work demonstrates that a mock‐Walker circulation is a useful generalization of radiative convective equilibrium that includes a large‐scale circulation. Plain Language Summary: Interactions between clouds, radiation, and dynamics all contribute to the large‐scale tropical motions and are fundamental to the Walker circulation. The Walker circulation is a loop consisting of surface winds toward the western tropical Pacific, strong upward motion and deep convection in that region, and the return eastward winds aloft that eventually sink toward the surface in the eastern Pacific basin. We focus on an idealization of the Walker circulation (a mock‐Walker circulation) in which the strong rising motion and deep convection is driven by a patch of warm sea surface temperature. Our results show that the response of the atmosphere to the radiative flux of energy depends strongly on the relative amount of clouds at different heights. It is further shown that our GCM‐like models are dominated by low‐clouds while our CRM‐like models are dominated by high‐clouds. This work also argues that an idealized Walker circulation is an excellent configuration with which to better understand the interactions between clouds, radiation, and circulation and to push the development of models forward. Models of mock‐Walker circulations represent an intermediate tier in a hierarchy of models between Earth‐like models and models of radiative convective equilibrium. Key Points: These cloud resolving simulations result in more upper‐level clouds, relative to low resolution simulations that have more low‐level cloudsHigh and low clouds interact differently with longwave radiation to increase or decrease precipitation, depending on the dominant cloud typeInteractions between clouds and radiation combined with parameterized convection shift the precipitation maximum away from the sea surface temperature maximum [ABSTRACT FROM AUTHOR]

Abstract: Satellite observation and general circulation model (GCM) studies suggest that precipitating ice makes nonnegligible contributions to the radiation balance of the Earth. However, in most GCMs, precipitating ice is diagnosed and its radiative effects are not taken into account. Here we examine the longwave radiative impact of precipitating ice using a global nonhydrostatic atmospheric model with a double‐moment cloud microphysics scheme. An off‐line radiation model is employed to determine cloud radiative effects according to the amount and altitude of each type of ice hydrometeor. Results show that the snow radiative effect reaches 2 W m−2 in the tropics, which is about half the value estimated by previous studies. This effect is strongly dependent on the vertical separation of ice categories and is partially generated by differences in terminal velocities, which are not represented in GCMs with diagnostic precipitating ice. Results from sensitivity experiments that artificially change the categories and altitudes of precipitating ice show that the simulated longwave heating profile and longwave radiation field are sensitive to the treatment of precipitating ice in models. This study emphasizes the importance of incorporating appropriate treatments for the radiative effects of precipitating ice in cloud and radiation schemes in GCMs in order to capture the cloud radiative effects of upper level clouds. [ABSTRACT FROM AUTHOR]

With general circulation models (GCMs) being increasingly used to explore extreme events over short temporal and small spatial scales, understanding how design choices in model configuration impact simulation results is critical. This research shows that the number of spontaneously generated tropical cyclones (TCs) in a version of the Community Atmosphere Model can be controlled by changing the coupling frequency between the dynamical core and physical parameterizations. More frequent coupling (i.e., shorter physics timesteps), even in the presence of an otherwise identical model, leads to large increases in TC activity. It is suggested that this arises due to competition within moist physics subroutines. Simulations with reduced physics timesteps preferentially eliminate instantaneous atmospheric instability via grid‐scale motions, even while producing mean climates similar to those with longer timesteps. These small‐scale variability increases lead to more tropical "seeds," which are converted to full‐fledged TCs. This behavior is confirmed through a set of sensitivity experiments and highlights the caution needed in studying and generalizing phenomena that depend on both resolved and sub‐grid scales in GCMs and the need for targeting physics‐dynamics coupling as a model improvement strategy. Plain Language Summary: Before an atmospheric model can produce data to be analyzed, a scientist must make a multitude of choices controlling subtle aspects of how the model runs. One such design choice is the "model timestep," or, how big the time increments are as the model marches forward in time. This work shows that the number of tropical cyclones in climate data is sensitive to this step size. As the step size is made smaller, the number of cyclones goes up, even if every other piece of the model code is kept identical. This occurs because changes in the timestep affect the amount of precipitation generated by different parts of the model and at which spatial scales it falls over. Key Points: The number of spontaneously generated tropical cyclones (TCs) in a free‐running general circulation model (GCM) can be controlled by physics timestepIn the Community Atmosphere Model, shorter timesteps promote higher TC frequencies due to upscale growth of near‐grid scale processesTropical cyclones frequencies at shorter timesteps can be reduced by increasing the amount of instability removed by the convective parameterization [ABSTRACT FROM AUTHOR]

A new spectral convection scheme, the stochastic unified convection scheme (stochastic UNICON), is implemented in a general circulation model. The global climate simulation using stochastic UNICON is evaluated and compared with UNICON, focusing on the simulation of the Madden–Julian oscillation (MJO). Stochastic UNICON extends the original UNICON by randomly sampling convective updrafts from the joint probability density function constructed at the near‐surface, generating a spectrum of convective updrafts in a physically based manner. The performances of UNICON and stochastic UNICON on simulating observed mean climates are comparable, while stochastic UNICON slightly reduces the mean bias of climate variables. For the simulation of intraseasonal variabilities, stochastic UNICON outperforms UNICON in many aspects. Stochastic UNICON improves the simulation of the intensity and propagation patterns of boreal winter MJO, which are too weakly simulated in UNICON. The coherency between MJO‐related convection and large‐scale circulation is also enhanced, which many climate models underestimate. The improvement of MJO simulation by stochastic UNICON is related to a better representation of the relationship between moisture and convection in the model. The increased frequency of shallow convection in stochastic UNICON leads to stronger moisture convergence that precedes convection activity peak and results in the more robust development of organized convection and more frequent intense precipitation. A precipitation budget analysis reveals that the moisture tendencies due to horizontal advection and convective process are consistently enhanced during MJO developing periods by stochastic UNICON. Plain Language Summary: The Madden–Julian oscillation (MJO) emerges in the tropics as a pattern of rainy periods followed by dry periods that fluctuates on weekly to monthly time scales. The MJO is important for various weather and climate phenomena but poorly represented in global climate models (GCMs). To simulate realistic MJO, convection needs to be realistically modeled. Recently, a new spectral convection scheme named stochastic unified convection scheme (stochastic UNICON) is developed based on the unified convection scheme (UNICON). Stochastic UNICON can represent different types of convective clouds in a model grid. The GCM with stochastic UNICON simulates better strength and propagation of winter MJO and better relationship between convection and large‐scale circulation compared to that of UNICON. The improvements come from a more realistic relationship between moisture and convection. The increased frequency of shallow convection in stochastic UNICON produces the condition for the MJO to develop more robustly. Key Points: The global simulation of the Madden–Julian oscillation (MJO) with a new spectral convection scheme named stochastic unified convection scheme (UNICON) is evaluatedStochastic UNICON improves various properties of the simulated MJO compared to the non‐stochastic versionThe increased shallow convection in stochastic UNICON leads to a stronger moistening tendency in MJO developing stage [ABSTRACT FROM AUTHOR]

Current conventional global climate models (GCMs) produce a weak increase in global-mean precipitation with anthropogenic warming in comparison with the lower tropospheric moisture increases. The motive of this study is to understand the differences in the hydrological sensitivity between two multiscale modeling frameworks (MMFs) that arise from the different treatments of turbulence and low clouds in order to aid to the understanding of the model spread among conventional GCMs. We compare the hydrological sensitivity and its energetic constraint from MMFs with (SPCAM-IPHOC) or without (SPCAM) an advanced higher-order turbulence closure. SPCAM-IPHOC simulates higher global hydrological sensitivity for the slow response but lower sensitivity for the fast response than SPCAM. Their differences are comparable to the spreads of conventional GCMs. The higher sensitivity in SPCAM-IPHOC is associated with the higher ratio of the changes in latent heating to those in net atmospheric radiative cooling, which is further related to a stronger decrease in the Bowen ratio with warming than in SPCAM. The higher sensitivity of cloud radiative cooling resulting from the lack of low clouds in SPCAM is another major factor in contributing to the lower precipitation sensitivity. The two MMFs differ greatly in the hydrological sensitivity over the tropical lands, where the simulated sensitivity of surface sensible heat fluxes to surface warming and CO2 increase in SPCAM-IPHOC is weaker than in SPCAM. The difference in divergences of dry static energy flux simulated by the two MMFs also contributes to the difference in land precipitation sensitivity between the two models. [ABSTRACT FROM AUTHOR]

The high computational cost of Global Climate Models (GCMs) is a problem that limits their use in many areas. Recently an inverse climate modeling (InvCM) method, which fixes the global mean sea surface temperature (SST) and evolves the CO2 mixing ratio to equilibrate climate, has been implemented in a cloud‐resolving model. In this article, we apply InvCM to ExoCAM GCM aquaplanet simulations, allowing the SST pattern to evolve while maintaining a fixed global‐mean SST. We find that InvCM produces the same climate as normal slab‐ocean simulations but converges an order of magnitude faster. We then use InvCM to calculate the equilibrium CO2 for SSTs ranging from 290 to 340 K at 1 K intervals and reproduce the large increase in climate sensitivity at an SST of about 315 K at much higher temperature resolution. The speedup provided by InvCM could be used to equilibrate GCMs at higher spatial resolution or to perform broader parameter space exploration in order to gain new insight into the climate system. Additionally, InvCM could be used to find unstable and hidden climate states, and to find climate states close to bifurcations such as the runaway greenhouse transition. Plain Language Summary: A large portion of the computational cost in climate simulation is spent in the initial equilibration phase before the model produces relevant output. This is because in normal Global Climate Model (GCM) simulations, you prescribe the atmospheric CO2 and allow the surface temperature to evolve until it reaches a roughly constant global‐mean value. In this article, we instead prescribe a global‐mean surface temperature and evolve the CO2 until the GCM is in energy balance. This leads to about a factor of ten decrease in equilibration time, which represents a considerable savings of computational resources. This allows us to investigate the change in climate sensitivity as the global‐mean temperature increases at high‐temperature resolution. Although the results we describe are for an idealized model set‐up, we believe that the method can be applied more generally. Additionally, the method could be used to investigate the climate response to given SST patterns and to find climate states that are unstable and/or hidden from normal GCM simulations. Key Points: We converge a GCM by varying the CO2 while keeping the global‐mean surface temperature fixed (the inverse climate modeling method)Inverse climate modeling converges about 10 times faster than normal slab‐ocean simulationsThe SST gradient response timescale is the bottleneck on the convergence rate of inverse climate modeling [ABSTRACT FROM AUTHOR]

Atmospheric gravity waves (GWs) span a broad range of length scales. As a result, the un‐resolved and under‐resolved GWs have to be represented using a sub‐grid scale (SGS) parameterization in general circulation models (GCMs). In recent years, machine learning (ML) techniques have emerged as novel methods for SGS modeling of climate processes. In the widely used approach of supervised (offline) learning, the true representation of the SGS terms have to be properly extracted from high‐fidelity data (e.g., GW‐resolving simulations). However, this is a non‐trivial task, and the quality of the ML‐based parameterization significantly hinges on the quality of these SGS terms. Here, we compare three methods to extract 3D GW fluxes and the resulting drag (Gravity Wave Drag [GWD]) from high‐resolution simulations: Helmholtz decomposition, and spatial filtering to compute the Reynolds stress and the full SGS stress. In addition to previous studies that focused only on vertical fluxes by GWs, we also quantify the SGS GWD due to lateral momentum fluxes. We build and utilize a library of tropical high‐resolution (Δx = 3 km) simulations using weather research and forecasting model. Results show that the SGS lateral momentum fluxes could have a significant contribution to the total GWD. Moreover, when estimating GWD due to lateral effects, interactions between the SGS and the resolved large‐scale flow need to be considered. The sensitivity of the results to different filter type and length scale (dependent on GCM resolution) is also explored to inform the scale‐awareness in the development of data‐driven parameterizations. Plain Language Summary: Gravity waves (GWs) present a challenge to climate prediction: waves on scales of O(1)–O(100) km can neither be systematically measured with conventional observational systems, nor properly represented (resolved) in operational climate models, which have a typical grid spacing on the order of 100 km. Therefore, in these climate models, small‐scale GWs must be parameterized, or estimated, based on the resolved (large‐scale) flow. The primary effects of these small‐scale waves on the resolved flow is the so‐called sub‐grid scale drag (Gravity Wave Drag [GWD]), resulting from the propagation and breaking of these waves. Existing GW parameterizations in general circulation models are all highly simplified; for example, they only account for vertical propagation of GWs. With growing computing power, a promising alternative approach is to use machine learning to develop data‐driven parameterizations. However, this requires to first generate reliable high‐resolution computer simulations and then extract GWD from these simulations. This study follows these steps, compares different extraction methods, and describes some challenges and pathways to make advances. Furthermore, our results suggest that the horizontal propagation of GWs should be included in parameterizations too, however, extra care is needed in order to extract the resulting GWD from high‐resolution data. Key Points: In a library of weather research and forecasting model simulations, we compare methods for estimating 3D gravity wave drag force that are un‐ and under‐resolved by general circulation modelsFor drag associated with vertical fluxes, different methods agree on time‐ and zonal‐mean but not on instantaneous spatiotemporal patternsDrag associated with horizontal fluxes is significant but is very sensitive to the estimation methodology [ABSTRACT FROM AUTHOR]

Previous work has found that as the surface warms the large‐scale tropical circulations weaken, convective anvil cloud fraction decreases, and atmospheric static stability increases. Circulation changes inevitably lead to changes in the humidity and cloud fields which influence the surface energetics. The exchange of mass between the boundary layer (BL) and the midtroposphere has also been shown to weaken in global climate models. What has remained less clear is how robust these changes in the circulation are to different representations of convection, clouds, and microphysics in numerical models. We use simulations from the Radiative‐Convective Equilibrium Model Intercomparison Project to investigate the interaction between overturning circulations, surface temperature, and atmospheric moisture. We analyze the underlying mechanisms of these relationships using a 21‐member model ensemble that includes both General Circulation Models and Cloud‐system Resolving Models. We find a large spread in the change of intensity of the overturning circulation. Both the range of the circulation intensity, and its change with warming can be explained by the range of the mean upward vertical velocity. There is also a consistent decrease in the exchange of mass between the BL and the midtroposphere. However, the magnitude of the decrease varies substantially due to the range of responses in both mean precipitation and mean precipitable water. We hypothesize based on these results that despite well understood thermodynamic constraints, there is still a considerable ability for the cloud fields and the precipitation efficiency to drive a substantial range of tropical convective responses to warming. Plain Language Summary: Tropical large‐scale overturning circulations are expected to weaken with warming. This weakening is the result of precipitation increasing at a slower rate than the atmospheric water vapor. Because precipitation and water vapor are important measures of how energy flows through the atmosphere it is important to understand how they will respond to a warming climate. We use two methods to calculate the change of the overturning circulation in 21 different simulations of the tropical atmosphere. This group of 21 models includes high resolution models that resolve cloud systems, and global models with grid‐spacing of about 100 km. We show that a weakening circulation that results from increasing atmospheric stability and increased water vapor is a robust result across most models. But across the group of models there is a large range of magnitudes in the response of the circulation to warming. This variability is well explained by the magnitude of the mean upward vertical velocity. Higher resolution models do not have a narrower range of responses. Narrowing this range of responses will depend on developing a better understanding of what drives the variations in atmospheric stability, surface fluxes of latent energy, and relative humidity. Key Points: The overturning tropical circulation weakens as the surface warms in the majority of radiative convective equilibrium models examinedInter‐model spread of circulation intensity and change of intensity are highly correlated with ascending velocity spread at 500 hPaVariability of both the clear‐sky heating and static stability result in large variations of the subsidence velocity [ABSTRACT FROM AUTHOR]

Gravity wave parameterizations contribute to uncertainties in middle atmosphere modeling. To investigate the potential for using machine learning to represent atmospheric gravity waves and the impact of implementing such schemes in a general circulation model (GCM), we train a random forest (RF) emulator on outputs from an existing complex parameterization scheme for orographic gravity wave drag (GWD). The performance of the RF emulator is then evaluated with a focus on stratospheric climatology and variability in climate simulations from the middle atmosphere resolving Beijing Climate Center Atmospheric Circulation Model. In offline tests, the predicted orographic GWD by the RF agrees remarkably well with the target GWD throughout the troposphere and the middle atmosphere. The RF emulator can reproduce the observed climatology of zonal‐mean wind and air temperature in the GCM simulation, as well as its target scheme. Compared to the target orographic GWD parameterization scheme, the RF emulator can reproduce the breakdown of the polar vortex in the Southern Hemisphere stratosphere. This study demonstrates the feasibility of using machine learning to emulate parameterized orographic GWD for modeling the stratosphere with a GCM. Plain Language Summary: Machine learning has been utilized to learn atmospheric physical parameterizations such as moist convection, cloud microphysics, radiation, nonorographic gravity waves, etc. However, it remains unknown whether it is feasible to apply machine learning algorithms to parameterize orographic gravity waves and how such schemes perform when interactively coupled in atmospheric general circulation models. Here we employ the widely used random forest algorithm to learn from an existing complex parameterization scheme for orographic gravity wave drag. Then, we implement the random forest scheme in a high‐top atmospheric general circulation model to test its performance. We find that the random forest scheme acts as well as the target parameterization scheme. This study demonstrates the potential of using machine learning to emulate parameterized orographic gravity waves for modeling the middle atmosphere. In future work, it is recommended that machine learning may be used to develop new parameterization schemes directly from observations or high‐resolution model outputs to overcome critical deficiencies in current orographic gravity wave parameterizations, helping to reduce the common biases in simulating the stratosphere with general circulation models. Key Points: Parameterized gravity wave drag from unresolved orography can be emulated with a random forestThe random forest emulator simulates vertical distributions of the zonal wind and air temperature quite well in a general circulation model (GCM)The GCM configured with the random forest scheme reproduces a reasonable breakdown of the austral polar vortex in the stratosphere [ABSTRACT FROM AUTHOR]

Representing subgrid variabilities of land surface processes and their upscaled effects is crucial for global climate modeling. Here, we implement a multiple atmosphere multiple land (MAML) framework in the superparamaterized version of E3SM (SP‐E3SM) to explicitly simulate the subgrid variabilities of land states and fluxes at cloud‐resolving scale and their interactions with atmosphere. Comparing to the standard SP‐E3SM in which all the atmospheric columns of the cloud resolving model embedded within the global atmospheric model grid interact with the same land surface (i.e., multiple atmosphere single land (MASL)), the impact of MAML on the strength of land‐atmosphere coupling is limited, partly because the current implementation mainly facilitates one‐way coupling between the cloud‐resolving model and the land surface model. Despite such limitation, MAML increases the surface latent heat flux at the expense of sensible heat flux, and increases precipitation in India, Amazon, and Central Africa, reducing the model dry bias compared to the standard SP‐E3SM. By employing a normalized gross moist stability (NGMS) diagnostic framework, we find that the increase in precipitation minus evaporation (P‐E) is primarily driven by the change in large‐scale moisture convergence, particularly by the increase of water vapor in the lower atmosphere, while the local effect of total surface energy flux plays a minor role in the P‐E change. More specifically, MAML changes the surface energy partitioning (evaporative fraction), increases the atmosphere water vapor, and further increases P‐E by decreasing the NGMS. Finally, future development in the MAML framework is discussed. Plain Language Summary: Land‐atmosphere interactions such as soil moisture‐precipitation feedback occur at a wide range of spatial scales. For example, spatial variability of surface fluxes such as radiation and precipitation can influence surface latent and sensible heat fluxes, which further influence turbulent mixing processes in the boundary layer and cloud and precipitation. Current global climate models (GCMs) have difficulties in representing land‐atmosphere interactions at scales smaller than the GCM grid (subgrid scales). To meet this challenge, we employ a novel framework, called multiple atmosphere multiple land (MAML), in a superparamaterized version of E3SM (SP‐E3SM) in which a cloud resolving model is embedded within each GCM grid to better resolve clouds and convection. MAML explicitly simulates the subgrid variabilities of land states and fluxes at scale of ∼1 km and provides feedback to the atmosphere. The use of MAML is shown to have a limited effect on the strength of land‐atmosphere coupling due to the current one‐way subgrid land‐atmosphere interaction setup. Despite the limited effect, MAML increases the rainfall in India, Amazon, and Central Africa, which reduces the dry bias in the model. Key Points: A multiple atmosphere multiple land (MAML) framework of land‐atmosphere coupling is implemented in the super‐parameterized E3SM (SP‐E3SM)MAML increases precipitation in India, Central Africa, and Amazon by increasing water vapor and hence large‐scale moisture convergenceMAML's impact on land‐atmosphere interactions is limited by the current setup of one‐way coupling on cloud‐resolving scales [ABSTRACT FROM AUTHOR]